A method, system, device and storage medium for intelligent comparison of bidding documents

Abstract

The application discloses a kind of intelligent comparison methods, systems, equipment and storage medium of bidding document.It is to be understood that the method includes: for the pretreatment of bidding document;Automatic segmentation independent tender terms, and classify tender terms, identify substantive terms and analyze;Segmentation bid document content, construct bid semantic unit set, generate semantic vector for each bid semantic unit, and retain the position information of each semantic unit;For each substantive clause, generate a semantic vector, and calculate the similarity between each tender clause vector and all bid semantic unit vectors, obtain the comprehensive semantic matching score;Use rule engine to make secondary determination on the matching result, obtain the rule engine determination result;For each tender substantive clause, based on the comprehensive semantic matching score and the rule engine determination result, generate response state and graded risk early warning, finally output structured comparison report.The application improves the reliability and efficiency of bid effectiveness review.

Description

Technical Field

[0001] This invention belongs to the field of bidding data processing and intelligent document comparison technology, specifically involving an intelligent comparison method for bidding documents based on natural language processing, semantic similarity calculation and rule engine. Background Technology

[0002] In the field of preparing and reviewing bidding documents, traditional methods mainly rely on manual comparison or automated tools based on string matching, which have problems such as low efficiency, easy omission of key clauses, and lack of business semantic judgment.

[0003] Existing technologies primarily employ manual comparison, document comparison tools based on string matching, AI-based plagiarism detection systems based on similarity calculation, and comparison techniques based on benchmark text databases. The limitations of existing solutions are analyzed in detail below.

[0004] (1) Manual comparison method Currently, many bidding agencies and review experts still rely on manual, page-by-page comparison of bidding documents and bids. This method has the following problems: Inefficient: A complete tender document is usually dozens to hundreds of pages long, containing technical parameters, commercial terms, qualification requirements and other contents. Reviewers need to check the response of the tender documents one by one, which is time-consuming and prone to fatigue and errors. Highly subjective: Different reviewers have inconsistent criteria for judging whether the same clause is a "valid response". For example, there is a lack of uniform standards for determining "equivalent substitution", which affects the fairness and repeatability of the review results.

[0005] Key clauses that are easily overlooked: Clauses marked with ★ or substantive clauses in bidding documents are usually scattered throughout the document and are very easy to overlook when reading manually. Once overlooked, the risk of bid rejection will not be detected in time.

[0006] (2) String matching-based document comparison tools To improve efficiency, some users use the document comparison function built into Word / WPS or general comparison tools such as Beyond Compare, but these solutions still have obvious shortcomings: Lack of semantic understanding: String matching tools can only recognize differences at the textual level and cannot understand semantic equivalence. For example, if a tender requires "CPU clock speed ≥ 2.8GHz" and the bid response is "2.9GHz", the string comparison will determine that the two are "different", while according to the bidding rules, they should be determined as "meets the requirements".

[0007] Unable to identify clause mapping relationships: For example, the requirements in Clause 3.2 of the tender document and the response in Clause 5.1 of the tender document may be related in content, but their positions do not correspond. General comparison tools cannot establish such cross-document clause mappings.

[0008] The output lacks business meaning: the comparison results are only text marks of "addition / deletion / modification", which cannot answer business questions such as "whether the clause is responsive", "whether it is a substantive requirement", and "whether it will lead to the rejection of the bid".

[0009] (3) AI plagiarism detection system based on similarity calculation In recent years, some companies have launched AI-based plagiarism detection systems for bidding scenarios (such as the Jinxiandai Bid Document Review Platform and the China Bidding Union Bid Document Similarity Analysis System). Their core function is to identify similar content between different bid documents to assist in identifying bid rigging. These systems mainly include a text plagiarism detection module (using NLP technology for semantic similarity calculation), an image plagiarism detection module (comparing the similarity of images such as certificates and seals), and a price analysis module (checking for regular differences in pricing). While these solutions have practical value in bid rigging detection scenarios, they are positioned as "horizontal comparisons between bid documents," serving only the purpose of identifying bid rigging. Existing plagiarism detection systems cannot be directly used for bidding response judgment. Specifically, they cannot distinguish between substantive clauses and general clauses (non-response to the former will lead to bid rejection, while the latter only affects the score), cannot output business semantic results such as "satisfied / unsatisfied," and cannot establish accurate cross-document clause-level mappings.

[0010] (4) Comparison techniques based on benchmark text databases Patent application CN202410990742.7 proposes a comparison scheme based on a benchmark text database. The process is as follows: compare the tender text with the benchmark database → obtain the target benchmark text → compare the bid text with the target benchmark text → perform the review operation. This scheme has the following shortcomings: Reliance on benchmark library construction: It is necessary to build a benchmark text library in advance that covers various bidding scenarios. The construction cost is high and it is difficult to cover the special requirements of emerging or niche industries.

[0011] Indirect matching has limited accuracy: When two rounds of matching are performed using a reference text as an intermediate bridge, the error accumulates step by step, and the matching accuracy is not as good as direct comparison.

[0012] Lack of a business rules engine: It still focuses on similarity calculation and has not integrated bidding and tendering business rules (such as the criteria for identifying substantive clauses and the rules for judging equivalent substitution).

[0013] In summary, the main problems with current bidding document processing technologies include: existing solutions either focus on "duplicate detection between bidding documents" or are limited to "string difference marking," resulting in a single comparison dimension and a lack of dedicated technical solutions for "response judgment between bidding documents and tender documents," and an inability to establish accurate cross-document clause-level mapping; based on string matching, semantic understanding is insufficient, and semantically equivalent content cannot be identified (such as "2.9GHz" and "≥2.8GHz"); due to the lack of risk classification, it is impossible to distinguish between substantive clauses and general clauses, lacks bid rejection warnings, and has weak rule adaptation capabilities; it usually outputs a similarity percentage rather than business semantics such as "satisfied / unsatisfied," because the results are uninterpretable.

[0014] Therefore, there is an urgent need for a more intelligent comparison method that is more in line with the business logic of bidding and tendering, so as to automatically identify substantive clauses, judge the response status item by item, and output graded risk reports, thereby reducing the rejection rate and improving the overall quality and efficiency of bidding and tendering activities. Summary of the Invention

[0015] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, system, device and storage medium for intelligent comparison of bidding documents.

[0016] According to a first aspect of the present invention, a method for intelligent comparison of bidding documents is provided. The method includes the following steps: The bidding documents are preprocessed to obtain structured bidding documents. The preprocessing includes text extraction, text cleaning, and tagging of structural information. For the structured tender documents, based on title level recognition and legal text pattern recognition, independent tender clauses are automatically segmented, and a lightweight classification model is used to classify the tender clauses to identify substantive clauses. Then, the substantive clauses are parsed to extract the constraint objects, constraint values ​​and ranges, and logical operators. The content of the tender document is segmented at the granular level of paragraphs, sentences, and table cells to construct a set of tender semantic units. Then, a semantic vector is generated for each tender semantic unit, and the location information of each semantic unit is retained for evidence tracing. For each of the aforementioned substantive clauses, a corresponding semantic vector is generated, and the similarity between each tender clause vector and all tender semantic unit vectors is calculated to obtain a comprehensive semantic matching score. Based on semantic matching, a rule engine is used to perform a second judgment on the matching results to obtain the rule engine judgment result; For each substantive clause in the tender, a response status and a graded risk warning are generated based on the comprehensive semantic matching score and the judgment result of the rule engine. Output structured comparison reports and evidence tracing information.

[0017] According to a second aspect of the present invention, a smart comparison system for bidding documents is provided. The system includes: Input and preprocessing module: used to preprocess the uploaded bidding documents to obtain structured bidding documents. The preprocessing includes text extraction, text cleaning and tagging of structural information. Parsing module: For the structured tender document, based on title level recognition and legal text pattern recognition, it automatically segments out independent tender clauses, and uses a lightweight classification model to classify the tender clauses to identify substantive clauses. Then, it parses the substantive clauses to extract constraint objects, constraint values ​​and ranges, and logical operators. Semantic Unit Segmentation Module: Segment the content of the tender document at the granularity of paragraphs, sentences, and table cells, construct a set of tender semantic units, generate a semantic vector for each tender semantic unit, and retain the location information of each semantic unit for evidence tracing; Semantic matching module: Generates a corresponding semantic vector for each of the aforementioned substantive clauses, and calculates the similarity between each tender clause vector and all tender semantic unit vectors to obtain a comprehensive semantic matching score; Rule Engine Module: Based on semantic matching, the rule engine performs a second judgment on the matching results to obtain the rule engine judgment result; Risk warning and analysis module: For each substantive clause in the tender, based on the comprehensive semantic matching score and the judgment result of the rule engine, a response status and graded risk warning are generated; Report generation module: Outputs structured comparison reports and evidence tracing information.

[0018] According to a third aspect of the present invention, a computer device is provided, including a memory and a processor, wherein a computer program capable of running on the processor is stored in the memory, wherein the processor executes the computer program to implement the steps of the above-described intelligent comparison method for bidding documents.

[0019] According to a fourth aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described intelligent comparison method for bidding documents.

[0020] Compared with existing technologies, the advantages of this invention are that the designed intelligent comparison scheme for bidding documents is positioned as a "vertical comparison between bidding documents and bid documents", which serves the review of bid validity. It can achieve a more intelligent comparison that is more in line with the business logic of bidding, and can automatically identify substantive clauses, judge the response status item by item, and output a graded risk report, thereby reducing the rejection rate and improving the overall quality and efficiency of bidding activities.

[0021] Other features and advantages of the invention will become clear from the following detailed description of exemplary embodiments of the invention with reference to the accompanying drawings. Attached Figure Description

[0022] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments of the invention and, together with their description, serve to explain the principles of the invention.

[0023] Figure 1 This is a flowchart of a method for intelligent comparison of bidding documents according to an embodiment of the present invention; Figure 2 This is an overall architecture diagram of a bidding document intelligent comparison system according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the document input and preprocessing module of an intelligent bidding document comparison system according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the automatic extraction and structured parsing module of the tender documents in the intelligent comparison system for tender documents according to an embodiment of the present invention. Figure 5 This is a schematic diagram of a response status generation and graded risk warning module according to an embodiment of the present invention. Detailed Implementation

[0024] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the invention.

[0025] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0026] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0027] In all the examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0028] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0029] This invention proposes an intelligent comparison method for bidding documents based on natural language processing, semantic similarity calculation, and a rule engine. In general, the method includes: first, converting unstructured bidding / tender texts into computable semantic units through structured parsing; then, combining a deep semantic matching model with a configurable bidding rule engine to achieve automatic response status determination and graded risk assessment for each substantive clause.

[0030] See Figure 1 As shown, the intelligent comparison method for bidding documents provided by this invention includes the following steps: Step 1: Preprocess the bidding documents to obtain structured bidding documents.

[0031] For example, it can obtain tender documents and bid documents, supporting formats such as PDF, Word, WPS, and OFD; it can perform text extraction and cleaning on the documents, including removing headers and footers, watermarks, non-text elements, redundant whitespace, and special control characters; it can retain and mark structural information such as tables, lists, and heading levels to obtain structured tender documents, providing a basis for subsequent clause segmentation.

[0032] Step 2: For the structured tender document, based on title level recognition and legal text pattern recognition, the independent tender clauses are automatically segmented, and a lightweight classification model is used to classify the tender clauses to identify substantive clauses. Then, the substantive clauses are parsed to extract the constraint objects, constraint values ​​and ranges, and logical operators.

[0033] Step 2 involves automatic extraction and structured parsing of substantive clauses, specifically including: Step 21: Based on title level recognition and legal text pattern recognition (such as "(I)", "1.", "1.1.", "★", etc.), automatically segment out independent bidding clauses; Step 22: Classify the terms using a lightweight classification model.

[0034] For example, lightweight classification models can employ various model architectures, such as those that have undergone lightweight transformation through model pruning, quantization compression, and knowledge distillation. Classification results include: substantive clauses (marked as key items, containing keywords such as "must," "must not," "★," and "※"), which are clauses that will lead to bid rejection if not responded to; general clauses, which only affect the score if not responded to; and descriptive / illustrative paragraphs (not included in the comparison). Step S23: Further analyze each substantive clause to extract the constraint object, constraint value and range, and logical operators.

[0035] For example, the constraints are "CPU clock speed", "after-sales service response time", "project manager qualification", etc.; the constraint values ​​and ranges are "≥2.8GHz", "within 24 hours", "ISO9001 certification", etc.; the logical operators are greater than / less than / include / not include / yes / no, etc.

[0036] Step 3: Segment the content of the tender document by paragraph, sentence, and table cell granularity, construct a set of tender semantic units, generate a semantic vector for each tender semantic unit, and retain the location information of each semantic unit for evidence tracing.

[0037] Step 3 involves semantic unit segmentation and vectorization, specifically including the following steps: Step 31: Segment the content of the tender document at the granular level of paragraphs, sentences, and table cells to construct a "tender semantic unit set"; Step 32: For each semantic unit, generate a semantic vector using a pre-trained Chinese domain NLP model (such as RoBERTa-wwm-ext, LawBERT, or a fine-tuned model for the bidding domain); Step 33: Simultaneously retain the location information (chapter, page number, paragraph ID) of each semantic unit for evidence tracing.

[0038] Step 4: Generate a corresponding semantic vector for each of the substantive clauses, and calculate the similarity between each tender clause vector and all tender semantic unit vectors to obtain a comprehensive semantic matching score.

[0039] Step 4 performs semantic similarity-based bidding-tender terms matching, which includes the following steps: Step 41: For each substantive clause of the tender, generate its semantic vector as well; Step 42: Calculate the cosine similarity or interactive matching score between each tender clause vector and all tender semantic unit vectors; Step 43: Select the Top-K bidding units with the highest similarity as candidate matching evidence.

[0040] For example, a threshold adaptive mechanism can be introduced: when the highest similarity is lower than the first threshold (e.g., 0.65), it is determined as "no matching content found"; when it is between the first threshold and the second threshold (e.g., 0.85), it is marked as "partial match / to be verified".

[0041] Step 5: Based on semantic matching, use the rule engine to perform a second judgment on the matching results to obtain the rule engine judgment result.

[0042] Step 5 executes the enhanced logic judgment of the rule engine. Specifically, based on semantic matching, the rule engine performs a second judgment on the matching results, including: Numerical and logical rules: If the tender terms include numerical constraints (such as "≥2.8GHz"), then extract the numerical values ​​from the tender matching units and compare them, outputting "satisfied / not satisfied"; Conditional rules: Handle conditional logic such as "if...then..." and "except for..."; Evidence chain combination rule: Allows response evidence for a single tender clause to be distributed across multiple bidding units (e.g., qualification requirements correspond to multiple certificates and documents), and the rule engine supports "AND / OR" logic combination judgment; Negative word processing: Identify negative structures such as "not possessing", "not provided", and "none" to prevent negative responses from being misinterpreted as positive responses.

[0043] Step 6: For each substantive clause of the tender, generate a response status and a graded risk warning based on the comprehensive semantic matching score and the judgment result of the rule engine.

[0044] Step 6 generates response status and graded risk warning. For example, for each substantive clause of the tender, the semantic matching score and the rule engine judgment result are combined to output one of the following five response statuses, see Table 1.

[0045] Table 1: Response Status Types and Descriptions

[0046] In addition, it can also output tiered risk warnings, such as: Red alert (risk of bid rejection): Substantive terms are judged to be "not met" or "not responded to"; Orange alert (point deduction / significant deviation risk): Substantive terms are judged to be "partially satisfied" or "pending verification"; Blue alert (General concerns): There are discrepancies in general terms; Green pass: Fully satisfied.

[0047] Step 7: Output a structured comparison report and evidence tracing information.

[0048] The process involves generating a comparison report and tracing evidence. For example, it outputs a structured comparison report, supporting formats such as HTML, PDF, Excel, and JSON. For each judgment result, it provides: the original text and location of the bidding terms, the original text and location (chapter, page number, etc.) of the matched bid response, semantic similarity score, the basis for the rule engine's judgment, risk warning level, and suggested actions (such as "It is recommended that the bidder supplement..." or "It is recommended that the review experts pay special attention to...").

[0049] In summary, this invention automatically identifies substantive clauses by integrating typesetting, keywords, and models, eliminating the need for manual annotation; it accurately matches industry terms, synonyms, and equivalent substitutions using a domain-fine-tuned semantic model; it combines semantic matching with a rule engine for fusion judgment, taking into account both semantic understanding and bidding business rules; and it enhances interpretability by outputting clause-level mapping and risk classification, directly serving bid evaluation and rejection warnings.

[0050] Accordingly, the present invention also provides an intelligent comparison system for bidding documents, used to implement one or more aspects of the above method. For example, see... Figure 2 As shown, the system includes: an input and preprocessing module: used to preprocess the uploaded bidding documents to obtain structured bidding documents, where preprocessing includes text extraction, text cleaning, and tagging structural information; a parsing module: used to automatically segment the structured bidding documents into independent bidding clauses based on title level recognition and legal text pattern recognition, and to classify the bidding clauses using a lightweight classification model to identify substantive clauses, and then parse the identified substantive clauses to extract constraint objects, constraint values ​​and ranges, and logical operators; and a semantic unit segmentation module (or bid vectorization module): used to segment the bid document content at the paragraph, sentence, and table cell granularity to construct bid semantic units. The system consists of a set of modules, which generate semantic vectors for each bidding semantic unit and retain the location information of each semantic unit for evidence tracing. A semantic matching module generates a corresponding semantic vector for each substantive clause and calculates the similarity between each bidding clause vector and all bidding semantic unit vectors to obtain a comprehensive semantic matching score. A rule engine module performs a secondary judgment on the matching results based on the semantic matching to obtain the rule engine judgment result. A risk warning analysis module generates a response status and graded risk warning for each bidding substantive clause based on the comprehensive semantic matching score and the rule engine judgment result. A report generation module outputs a structured comparison report and evidence tracing information.

[0051] The file input and preprocessing module supports multi-format document parsing, text cleaning, and preservation of structural information. (See also...) Figure 3 As shown. The parsing module implements clause extraction and parsing, segments and categorizes clauses, and extracts constraint objects, numerical ranges, and logical operators, such as... Figure 4 As shown. The semantic unit segmentation module segments the bidding content into semantic units and generates vectors, while retaining positional information. The semantic matching module calculates the similarity between the bidding terms and the bidding units, for example, by filtering matching evidence through a dual threshold mechanism. The rule engine module performs a secondary, precise judgment on the matching results based on numerical logic, conditional rules, and negative word processing. The risk warning analysis module can output five levels of status, such as "satisfied / not satisfied," and mark four levels of risk warnings: red, orange, blue, and green. Figure 5 As shown. The report generation module is used to generate structured reports, providing original source information, judgment criteria, and operational suggestions.

[0052] It should be noted that, without departing from the spirit and scope of this invention, those skilled in the art can make appropriate changes or modifications to the above embodiments. For example, the implementation of bidding document comparison can also be achieved without using a deep learning semantic model, but solely through regular expressions, keyword dictionaries, numerical templates, and clause-level rules to complete document parsing, keyword matching, and rule determination. This approach can also automate the comparison of bidding documents, but it has the following limitations: its ability to identify synonyms, equivalent substitutions, and ambiguous expressions is relatively weak, its matching accuracy is relatively low, and it is difficult to handle complex or non-standardized bidding response content. Therefore, this approach is only suitable for bidding document comparison scenarios with simple structures, standardized content, and a high degree of templateization.

[0053] In summary, compared with the prior art, the present invention has the following advantages: (1) Adopting a domain-fine-tuned semantic model, it accurately identifies industry terms, synonyms and equivalent substitutions, with high semantic matching accuracy, thus solving the limitations of traditional string matching.

[0054] (2) The integration of layout, keywords and model to automatically identify substantive clauses eliminates the need for manual annotation, accurately identifies risk items for rejection, and avoids omission of key clauses.

[0055] (3) The fusion architecture of semantic matching and rule engine takes into account both semantic understanding and bidding business rules, which greatly improves the accuracy and reliability of comparison.

[0056] (4) Output graded risks and interpretable business results to directly support the evaluation decision, significantly improve the efficiency and fairness of the review and reduce the risk of rejection.

[0057] This invention can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of the invention.

[0058] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0059] Various aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0060] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0061] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0062] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions. It will be known to those skilled in the art that implementation in hardware, implementation in software, and implementation using a combination of software and hardware are equivalent.

[0063] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, and are not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein. The scope of the invention is defined by the appended claims.

Claims

1. A method for intelligent comparison of bidding documents, comprising the following steps: The bidding documents are preprocessed to obtain structured bidding documents. The preprocessing includes text extraction, text cleaning, and tagging of structural information. For the structured tender documents, based on title level recognition and legal text pattern recognition, independent tender clauses are automatically segmented, and a lightweight classification model is used to classify the tender clauses to identify substantive clauses. Then, the substantive clauses are parsed to extract the constraint objects, constraint values ​​and ranges, and logical operators. The content of the tender document is segmented at the granular level of paragraphs, sentences, and table cells to construct a set of tender semantic units. Then, a semantic vector is generated for each tender semantic unit, and the location information of each semantic unit is retained for evidence tracing. For each of the aforementioned substantive clauses, a corresponding semantic vector is generated, and the similarity between each tender clause vector and all tender semantic unit vectors is calculated to obtain a comprehensive semantic matching score. Based on semantic matching, a rule engine is used to perform a second judgment on the matching results to obtain the rule engine judgment result; For each substantive clause in the tender, a response status and a graded risk warning are generated based on the comprehensive semantic matching score and the judgment result of the rule engine. Output structured comparison reports and evidence tracing information.

2. The method according to claim 1, characterized in that, For each of the bidding semantic units, a pre-trained Chinese domain NLP model is used to generate the corresponding semantic vector.

3. The method according to claim 1, characterized in that, The location information for each semantic unit includes chapter, page number, and paragraph ID.

4. The method according to claim 1, characterized in that, The similarity is cosine similarity or interactive matching score.

5. The method according to claim 1, characterized in that, The secondary determination of the matching results using a rule engine includes: Numerical and logical rules: If the bidding terms contain numerical constraints, extract the values ​​from the bid matching units and compare them, outputting whether the conditions are met or not. Conditional rules: handle conditional logic; Evidence chain combination rule: Allows response evidence for a single tender clause to be distributed across multiple tender units; the rule engine supports AND / OR logic combination judgment. Negation word processing: Identifying negation structures.

6. The method according to claim 1, characterized in that, The response status includes: Completely satisfies: semantic equivalence and complete numerical logic conformity; Partially satisfies: the main semantics match, but the details are not completely consistent; Not satisfied: semantically opposite or key condition missing; Pending verification: Similarity is below the set threshold or the evidence is ambiguous; No response: No matching content found; The tiered risk warning system includes: Risk of bid rejection: Substantive terms may be deemed "unmet" or "unresponsive"; Risk of deduction: Substantive terms may be judged as "partially satisfied" or "pending verification"; General concerns: There are differences in the general terms; Pass: Fully satisfied.

7. The method according to claim 1, characterized in that, The structured comparison report includes: for each judgment result, the original text and location of the bidding terms, the original text and location of the matched bid response, the semantic similarity score, the basis for the rule engine's judgment, the risk warning level, and suggested actions.

8. A smart comparison system for bidding documents, comprising: Input and preprocessing module: used to preprocess the uploaded bidding documents to obtain structured bidding documents. The preprocessing includes text extraction, text cleaning and tagging of structural information. Parsing module: For the structured tender document, based on title level recognition and legal text pattern recognition, it automatically segments out independent tender clauses, and uses a lightweight classification model to classify the tender clauses to identify substantive clauses. Then, it parses the substantive clauses to extract constraint objects, constraint values ​​and ranges, and logical operators. Semantic Unit Segmentation Module: Segment the content of the tender document at the granularity of paragraphs, sentences, and table cells, construct a set of tender semantic units, generate a semantic vector for each tender semantic unit, and retain the location information of each semantic unit for evidence tracing; Semantic matching module: Generates a corresponding semantic vector for each of the aforementioned substantive clauses, and calculates the similarity between each tender clause vector and all tender semantic unit vectors to obtain a comprehensive semantic matching score; Rule Engine Module: Based on semantic matching, the rule engine performs a second judgment on the matching results to obtain the rule engine judgment result; Risk warning and analysis module: For each substantive clause in the tender, based on the comprehensive semantic matching score and the judgment result of the rule engine, a response status and graded risk warning are generated; Report generation module: Outputs structured comparison reports and evidence tracing information.

9. A computer device comprising a memory and a processor, wherein a computer program capable of running on the processor is stored in the memory, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images